Machine learning is one subfield of artificial intelligence (AI). Some AI models have been developed using risk thresholds based on the Problem Gambling Severity Index (PGSI) to identify people experiencing harms. But the performance of AI models may worsen over time. The researchers developed two machine learning models in 2019. These models sought to detect people at a high risk of experiencing gambling harms in the last year (PGSI scores = 8+) and people at moderate-to-high risk (PGSI scores = 5+).
This study assessed the stability of the models developed in 2019 using data collected in 2022. Participants were 11,258 adult account holders (18+ years) on the Canadian gambling website lotoquebec.com. Participants completed the PGSI and agreed to provide their past-year account data to the researchers. The researchers noted that there were some changes in the models’ performance over time. After some minor adjustments made to the models’ decision thresholds, the two models were still able to correctly identify people at risk of harms using indicators of online gambling behaviours.
